Research map / Vision-language and multimodal / CLIP and zero-shot
AGO: Adaptive Grounding for Open World 3D Occupancy Prediction
Peizheng Li, Shuxiao Ding, You Zhou, Qingwen Zhang, Onat Inak, Larissa Triess, Niklas Hanselmann, Marius Cordts, Andreas Zell
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use adaptive grounding (modality adapter plus grounding) instead of cosine alignment with image embeddings (ablation) | AGO · Occ3D-nuScenes (self-supervised) | mIoU: 10.28 -> 19.23 | Improved |
| use adaptive grounding instead of pure grounding (ablation) | AGO · Occ3D-nuScenes (self-supervised) | mIoU: 19.08 -> 19.23 | No clear effect |
| use adaptive grounding instead of pure grounding on unknown classes (ablation) | AGO · Occ3D-nuScenes (open-world, pretraining) | unknown mIoU: 0.0 -> 3.6 | Improved |
| add alignment loss on the same embeddings as grounding (ablation) | AGO (Gro.) · Occ3D-nuScenes (self-supervised) | mIoU: 19.08 -> 18.89 | Hurt |
| use adaptive grounding instead of grounding plus alignment on the same embedding (ablation) | AGO · Occ3D-nuScenes (open-world, pretraining) | known mIoU: 18.3 -> 22.1 | Improved |
| add open world identifier with min entropy criterion (ablation) | AGO · Occ3D-nuScenes (open-world, pretraining) | unknown mIoU: 1.1 -> 3.6 | Improved |
| add open world identifier with min entropy criterion (ablation) | AGO · Occ3D-nuScenes (open-world, zero-shot) | unknown mIoU: 1.6 -> 3.2 | Improved |
| use min entropy instead of max confidence criterion (ablation) | AGO · Occ3D-nuScenes (open-world, pretraining) | unknown mIoU: 3.1 -> 3.6 | Improved |
| add occupancy loss L_Occ (ablation) | AGO · Occ3D-nuScenes (self-supervised) | mIoU: 18.01 -> 18.97 | Improved |
| add occupancy loss L_Occ (ablation) | AGO · Occ3D-nuScenes (self-supervised) | IoU: 53.10 -> 55.18 | Improved |
| add random noise prompts as negative samples (ablation) | AGO · Occ3D-nuScenes (self-supervised) | mIoU: 18.97 -> 19.23 | Improved |
| use ResNet-50 image backbone instead of ResNet-101 (ablation) | AGO · Occ3D-nuScenes (self-supervised) | mIoU: 19.23 -> 15.23 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- POP-3D — POP-3D: Open-Vocabulary 3D Occupancy Prediction from Images (NeurIPS 2023) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LOMA: Language-assisted Semantic Occupancy Network via Triplane Mamba (AAAI 2025 · similarity 0.91)
- OccMamba: Semantic Occupancy Prediction with State Space Models (CVPR 2025 · similarity 0.90)
- LowRankOcc: Tensor Decomposition and Low-Rank Recovery for Vision-based 3D Semantic Occupancy Prediction (CVPR 2024 · similarity 0.90)
- AdaOcc: Adaptive 3D Occupancy Prediction for Embodied Tasks (NeurIPS 2026 · similarity 0.89)
- OctOcc: High-Resolution 3D Occupancy Prediction with Octree (AAAI 2024 · similarity 0.88)
- Achieving Speed-Accuracy Balance in Vision-based 3D Occupancy Prediction via Geometric-Semantic Disentanglement (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
